Hi, getting this message during a session :

process memory usage is high

.

Thanks for reaching out, Sailendra. I'm an AI assistant and I'm here to help. Give me a moment while I research this topic.

Hi, getting this message during a session :

process memory usage is high

Thanks — I see the

process memory usage is high

message you’re seeing during a session. I can help narrow this down and recommend next steps. What this typically means • Your LiveKit agent process is using a large amount of RAM for the current workload. Agents that use features like enhanced noise cancellation, speech-to-text (STT), or voice activity detection (VAD) require more memory and CPU. • As a practical baseline, LiveKit recommends about 4 CPU cores and 8 GB RAM per agent for many voice-AI workloads; in a load test, a 4-core/8GB machine with 30 concurrent sessions peaked around ~2.8 GB memory usage. Memory needs scale with the number of concurrent sessions and which agent components you enable. What I need from you to diagnose Please share any of the following that you can: • Deployment type: LiveKit Cloud or self-hosted. • Agent configuration / enabled features (e.g., STT, TTS, enhanced noise cancellation, Silero VAD). • LiveKit server or agent version. • Approximate number of concurrent sessions on that agent when the warning appears. • Whether memory usage steadily increases over time (possible leak) or spikes only under load. • Room ID / participant ID and logs or screenshots of the warning (redact any sensitive info). Immediate steps you can try • If self-hosted, check agent host metrics (memory and CPU) around the time of the warning to confirm whether usage is sustained or transient. • Reduce or disable optional, memory-heavy features (STT, advanced noise cancellation, etc.) temporarily to see if memory drops. • Restart the agent process to see if memory usage returns to normal (good for a quick mitigation, but not a root-cause fix). • If you have autoscaling, increase agent capacity or add another agent before the host hits capacity. If you share the details above (deployment type, enabled features, concurrent session count, and any logs), I’ll give concrete next steps tailored to your environment — for example, recommended resource sizing, configuration adjustments, or targeted troubleshooting commands. Sources: Self-hosted deployments | LiveKit Documentation

• Deployment type: LiveKit Cloud. • STT, TTS, enhanced noise cancellation, Silero VAD: enabled. • Approximate number of concurrent sessions: 3 • 4 - 7 participants

Thanks — this is helpful. You’re on LiveKit Cloud with memory-heavy features enabled (STT, TTS, enhanced noise cancellation, Silero VAD) and seeing the

process memory usage is high

warning during sessions with 3 concurrent agent sessions and 4–7 participants. I’ll explain what’s likely happening and give clear, actionable steps you can take now. Why this is happening • Those features (STT, TTS, enhanced noise cancellation, and Silero VAD) increase per-session memory usage. Even with only 3 concurrent sessions, the combined load from audio processing and ML models can push agent memory toward the host limits. • On typical deployments, a recommended baseline is around 4 CPU cores and 8 GB RAM per agent for voice-AI workloads; memory usage scales with feature set and number of simultaneous participants. Seeing a high-memory warning with your configuration is expected when multiple heavy features run together. What to check right now • Observe whether memory usage is sustained or temporary (spike vs. steady growth). A steady increase over time suggests a leak; spikes suggest load peaks from processing bursts. • Note whether the warning correlates with specific actions (e.g., starting transcription, joining a new participant, or starting screen/audio shares). • Gather logs around the warning time (agent logs and session logs). If you can paste the relevant log lines here (redact any sensitive IDs), I can interpret them. Immediate mitigations you can apply • Restart the affected agent process to immediately reduce memory usage (short-term fix for spikes). • Temporarily disable one or more memory-heavy components (for example, disable enhanced noise cancellation or Silero VAD) and observe whether memory usage drops. • Stagger or limit concurrent jobs that run STT/TTS-heavy workflows where possible. Recommended next steps (medium-term) 1. Scale agent capacity: ◦ Increase the agent instance size (more RAM) or add additional agents so sessions are spread across more processes. ◦ Autoscale earlier than the host limit so you avoid hitting high-memory warnings under load. 1. Tune feature usage: ◦ If full-time STT/TTS isn’t required for every session or participant, enable them selectively when needed. ◦ Use lighter-weight VAD/noise-suppression options if available. 1. Monitor and profile: ◦ Enable continuous monitoring of agent memory and CPU to detect trends. ◦ If memory grows over time even without load spikes, collect a memory profile or heap snapshot and logs for the time range when growth occurs. What I can help with next • If you paste the relevant agent log lines (surrounding the

process memory usage is high

warning) I’ll analyze them and point to likely causes. • I can recommend specific resource sizing for LiveKit Cloud given your feature set and expected concurrency. • If you want, tell me which feature you can temporarily disable first and I’ll outline exactly how to do that and what metrics to watch. If you share logs or tell me which mitigation you want to try first, I’ll guide you step-by-step.